The browser reads the file
Your recording is opened from disk and decoded by the browser's own audio engine. It is not uploaded.
Local audio transcription
Transcribe audio locally for free, without uploading it
EarScribe decodes and transcribes your recording inside the browser tab. The file is never sent to a server; the only large download is the speech model, which your browser caches. This page explains what stays on your device, what the page does send, and how to verify it yourself.
- The recording never leaves your device
- Nothing to install and no account
- Checkable in your browser's developer tools
Drop an audio file or click to browse
MP3, WAV, M4A, OGG, FLAC, WebM — practical length depends on device memory
Review the transcript before you export
Free local transcription, step by step
The model runs in a worker
Whisper runs in a background Web Worker through Transformers.js and ONNX Runtime Web, on WebGPU or the CPU.
The result stays in the tab
The transcript appears on the page, and a draft is kept in this browser's storage until you clear it.
transcribe audio locally
How local transcription works, and how to verify it
Most online transcription services upload your file to their servers. EarScribe takes the opposite approach: the model comes to your browser, and the recording stays where it is.
What stays on your device and what doesn't
Stays on the device: the audio file, the decoded audio, the transcript, your edits and exported files. Goes over the network: the page itself, the Whisper model files on first use, aggregate analytics events such as a finished transcription (without file names or text), and Google's advertising scripts, as described in the privacy policy.
The model download, roughly 80 MB to 800 MB depending on the model, is cached by the browser, so later sessions on the same device skip it.
Verify it with the Network panel
In Chrome or Edge, press F12 (Option+Command+I on a Mac) and open the Network tab; Safari needs its Develop menu turned on first. Clear the list, choose a file and transcribe it. On a first run you'll see requests for model files, but none whose size matches your recording.
Sorting the list by size makes this obvious. For a stricter test, transcribe once so the model is cached, reload the page, and watch the network while a second file is processed.
When local processing is the right choice
Local transcription suits recordings you can't or don't want to hand to a third party: confidential interviews, internal meetings, research sessions whose consent forms rule out outside services, or personal notes.
It doesn't make a device secure on its own. Malware, browser extensions and shared accounts can still see what you open, so give the transcript the same care as the recording.
Review the transcript before you export
Checklist for sensitive recordings
- 1
Profile
Use a browser profile without unnecessary extensions.
- 2
Verify
Watch the Network panel once if you need proof.
- 3
Clean up
Clear site data to remove saved drafts on shared devices.
- 4
Exports
Store exported transcripts where your policies allow.
Honest limits
- EarScribe is a web page, not an installed offline app; it needs a connection to load and to download the model.
- The page also loads analytics and advertising scripts, described in the privacy policy; EarScribe never sends them your audio or transcript.
Local doesn't mean careless
Browser extensions can read what a page shows. Use a clean browser profile for sensitive recordings.
Exported files and saved drafts live on this device. On a shared computer, delete them and clear the site's data.
Follow your organization's rules for confidential material, even when nothing is uploaded.
Questions about this workflow
Is my audio uploaded anywhere?
No. The file is read and processed in the browser tab, and EarScribe's servers never receive the recording or the transcript.
What does the page download or send?
The page's own code, the Whisper model the first time you use it (through EarScribe's model proxy), aggregate Plausible analytics events without file names or transcript text, and Google's advertising scripts. None of these requests contain your audio.
Does it work offline?
Not as an installed offline app. You need a connection to load the page and to download the model the first time, but the transcription itself never depends on a server.
How can I check this myself?
Open your browser's developer tools, go to the Network tab and transcribe a file. You'll see the model files on the first run, and no request that carries your recording.